The environmental representativeness effect in species distribution model evaluation without absence data

Resumen

In species distribution modeling (SDM), the environmental representativeness effect hinders the comparison and generalization of discrimination statistics, as their values are context-dependent and may vary without reflecting actual differences in model accuracy. To address this issue, a harmonization approach based on the uniform distribution of suitability values has been proposed, demonstrating effectiveness when true absence data are available. However, in most cases, models solely rely on presence records and use background points instead of true absences, posing additional validation challenges. This study simulates habitat suitability and presence-absence data, and evaluates the robustness of several validation indices: background-based AUC (AUC_b), its harmonized version (uAUC_b), and two variations of the Boyce index. The results show that the Boyce index remains unaffected by the representativeness effect, whereas AUC_b varies across scenarios, confirming the influence of the representativeness effect on SDM results with background data. Harmonization through uAUC_b successfully makes values comparable, but its reliability depends on sample size, requiring at least 100 presences and 10000 background points.